Regulated intramembrane proteolysis is an evolutionarily conserved mechanism by which membrane-anchored bioactive molecules are released from cellular membranes. In eukaryotic cells, intramembrane proteases are found in different cellular organelles ranging from the endosomal system to mitochondria and chloroplasts. These proteases function in diverse processes such as transcription control, regulated growth factor secretion and recently even a role in the control of mitophagy has been suggested. Genomic annotation has predicted 13 different intramembrane proteases in humans. Apart from few studied examples, very little is known about their function. This review describes emerging principles of how intramembrane proteases contribute to the regulation of cellular protein trafficking in eukaryotic cells and raises the important question of how their activity is controlled.
Proteases are a potent class of enzymes that catalyze the hydrolysis of peptide bonds. Because of their irreversible effects, they introduce directional changes to protein networks, which urges for a tight regulation. Controlled endoproteolysis plays a prominent role in many biological pathways, but it is equally important to control cellular protein homeostasis. Malfunction of this so-called proteostasis network has been linked to various human diseases and molecular aging and therefore proteases are among the most promising drug targets that are currently investigated.
Humans have around 500 different proteases, which form one of the largest enzyme classes and localize to various organelles of the cell. Despite a strict requirement for water, several proteases have evolved the ability to cleave peptides in the hydrophobic environment of the membrane. The substrates of these intramembrane proteases are commonly transmembrane (TM) proteins fulfilling a wide range of biologically important functions. Intramembrane proteases and their substrates have also been implicated in human diseases such as Alzheimer's disease (AD), Parkinson's disease (PD), type 2 diabetes mellitus, as well as infections by viruses and pathogenic microorganisms. Recent progress in the understanding of the structure and catalytic mechanism of these unusual proteases has been reviewed extensively (1–4). This review describes key examples of how intramembrane proteases release membrane-tethered factors, providing the trigger for their relocation to a new destination or initiating protein degradation. Moreover, I frame fundamental questions as to how regulated intramembrane proteolysis (RIP) is restricted to specific cell compartments, and as to the mechanism by which cleavage specificity is ensured in the complex biomembrane system.
Overview of Intramembrane Proteolysis
In most species that have been analyzed, about one third of all genes encode membrane integral proteins (5). Their fate is fundamentally different from soluble proteins because of the biophysical properties of cellular membranes (6). For most of their life, TM proteins are stably anchored in the lipid bilayer, commonly by hydrophobic membrane-spanning helices, and these proteins generally face an energetic barrier for their release. Besides energy-dependent dislocation of TM domains by AAA-type ATPases (7), intramembrane proteolysis is an efficient way to release membrane-tethered proteins. In the simplest case, RIP generates two fragments and releases one to each side of the membrane (as outlined in Figure 1 and the following paragraphs). This can trigger relocation of a substrate domain, liberate a dormant protein activity or initiate protein degradation. Bioactive fragments that are released into the cytosol are commonly degraded unless they are stably bound to a protein or enter a spatial niche such as the nucleus that is believed to be free of most proteases (8). The emerging picture is that the turnover rate of cytosolic cleavage fragments controls the duration of transient signals (9) and aggregation-prone hydrophobic peptides are rapidly removed from the cell (10). Contrary to the cytosolic fragments, cleavage products released into the exterior or luminal space are more stable and are commonly secreted (11,12).
While first hypothesized in the 80s to explain an unexpected cut in the TM domain of the β-amyloid precursor protein (APP) (13), it took a while until the concept of intramembrane proteolysis was fully accepted. Initially identified as exotic exceptions, now several intramembrane proteases have been annotated in all eukaryotic genomes (14–20), indicating that RIP is a universal cellular process. Intramembrane proteases are key players in diverse cellular processes ranging from transcriptional control, cellular signaling to the control of mitochondrial membrane dynamics. For some predicted intramembrane proteases, no substrate has been identified so far, suggesting that more RIP pathways wait to be discovered. Like for soluble proteases, the main classification of intramembrane proteases is based on the catalytic motifs, which are buried in the membrane forming a water-containing active site (see Table 1 and the following paragraphs). In humans, 13 different intramembrane proteases are known and with the exception of peroxisomes they are found in all membrane-enclosed cellular organelles (12,19,21–26). With metallo-, aspartyl- and serine proteases, members of three of the four main mechanistic groups are known; however, no intramembrane cysteine protease has been identified yet. Alternatively, intramembrane proteases can be grouped according to the topology of their substrates (Figure 1). The currently accepted view is that a particular intramembrane protease shows a high preference for one substrate orientation—it either cleaves type I membrane proteins (N-terminus outside and C-terminus in the cytosol) or type II-oriented TM segments (N-terminus inside and C-terminus outside).
Table 1. Different families and subgroups of intramembrane proteases in human and the budding yeast S. cerevisiae
SPPL2a, SPPL2b, SPPL2c
RHBDL1, RHBDL2, RHBDL3
Release of Membrane-Tethered Transcription Factors by the S2P Metalloprotease
The first intramembrane protease known was the site 2-protease (S2P). It was identified by complementation cloning of factors that activate the sterol regulatory element-binding protein (SREBP), a conserved transcription factor involved in the feedback regulation of sterol and lipid biosynthesis (27). The unexpected outcome was that the transcriptional active fragment of SREBP, which is synthesized as a hairpin TM precursor, is liberated from the membrane by two sequential cuts (Figure 2A). First, S1P, a subtilisin-related serine protease, cleaves within the luminal loop (28). Second, S2P cleaves in the TM domain of the N-terminal fragment thereby releasing the transcriptionally active SREBP domain that translocates to the nucleus to activate the expression of SREBP target genes (27,29). Another key function of S2P in animals is the activation of the membrane-bound activating transcription factor 6 (ATF6), which is a type II membrane protein that acts as a key activator of the endoplasmic reticulum (ER) unfolded protein response (UPR) (30). Common to these RIP pathways is that the S2P-catalyzed intramembrane cut of a type II-oriented TM segment has to be preceded by S1P substrate tailoring (Figure 2A). This sequential action of two proteases ensures that S2P, which shows only limited sequence specificity, acts in a controlled manner (31).
The S2P family members are metalloproteases with conserved His-Glu-x-x-His and Asp-Gly active site motifs in two separate TM helices (Figure 2A) (27,32). These two motifs bind the prosthetic zinc ion required for the activation of the catalytic water. S2P is conserved in metazoans and some fungal genomes, but was lost in the budding yeast Saccaromyces cerevisiae (Table 1). Plants have evolved several copies of S2P homologues that show differential localization. Known functions of Arabidopsis thaliana S2P homologues range from activation of the UPR by cleaving ER-resident transcription factors (33) to a role in plant development by cleaving an unknown factor of the inner chloroplast membrane (34).
GxGD-Type Aspartyl Proteases Scatter Across the Secretory Pathway
The by far largest and most heterogeneous group of intramembrane proteases is formed by the GxGD-type aspartyl proteases that share the conserved Tyr-Asp and Gly-x-Gly-Asp active site motifs in adjacent TM segments (Figure 2B,C) (11,19). In analogy to classical proteases, it seems probable that the conserved aspartate residues activate water for the nucleophilic attack of the scissile peptide bond. Function and localization of GxGD-type intramembrane proteases are quite diverse, and depending on the orientation of the active site TM helices in the membrane, they are either specific for type I or type II membrane protein substrates (Figure 1).
γ-Secretase: From signaling to disposal of type I TM stubs
The best-characterized GxGD-protease is γ-secretase. Its active subunit, termed presenilin, was identified in a screen for genes that are linked to early-onset familial AD (14,15). Work by a number of laboratories subsequently showed that presenilin promotes the cleavage of APP (11,35). Again, the intramembrane cut has to be preceded by shedding of the APP ectodomain close to the membrane (Figure 2B) (36). The subsequent presenilin-catalyzed cleavage leads to the release of short hydrophobic N-terminal peptides toward the exterior side of the membrane, of which Aβ is the most renowned (11). In a dual cleavage mechanism, presenilin also cleaves near the inner membrane surface at the ε-site releasing the C-terminal intracellular domain of APP (37). Whereas this dual cleavage mechanism and the biological function of the APP intracellular fragment are not completely understood, the secreted amyloidogenic Aβ peptide represents a major risk factor for AD (for reviews see (4,23,36)).
In addition to its role in AD, work in Caenorhabditis elegans revealed a conserved function of γ-secretase in the Notch signaling pathway (16). A key step in Notch activation is the proteolytic release of the intracellular domain, which translocates to the nucleus where it acts as transcriptional activator of genes involved in patterning processes during development (38). Meanwhile, more than 60 γ-secretase substrates have been identified (23). The emerging picture is that γ-secretase-catalyzed cleavage does not depend critically on the recognition of any primary sequence motifs but instead on the size of the extracellular domain (39). This suggests that besides its specific function in signaling, γ-secretase participates in the proteostasis network by clearing TM protein remnants from the plasma membrane (40).
γ-secretase differs from other intramembrane proteases by its assembly within the membrane. Whereas no other membrane protease needs cofactors, additional non-catalytic subunits are essential for γ-secretase activity (41–43). The active subunit presenilin associates with the type I membrane protein nicastrin, the bitopic membrane protein Pen-2 and Aph-1 that spans the membrane seven times (for reviews see 4,23). The four γ-secretase subunits exist in metazoans and plants, but are missing in fungi. Mammals have two presenilin homologues and several Aph-1 isoforms (three in humans and four in rodents), suggesting that different complexes exist (44). Intriguingly, a recent analysis of individual and combined knockouts of Aph-1 genes in mice points to the intriguing possibility that the substrate specificity of γ-secretase depends on the Aph-1 isoform (45).
SPP-type proteases cleave type II TM domains in various different cellular organelles
An activity-based cross-linking approach identified another GxGD-type protease (19). The protease was termed signal peptide peptidase (SPP), as it removes signal sequences from the ER membrane. Upon targeting of nascent polypeptide chains to the ER, signal peptides are inserted into the membrane and then cleaved from their precursor protein by signal peptidase to become further processed by SPP (Figure 2C). Like for γ-secretase and S2P, first substrate peptides have to be liberated from the luminal portion of the nascent chain in order to be cleaved by SPP (46). The majority of signal peptide fragments generated by SPP are probably immediately degraded. In some cases, however, signal sequence-derived fragments serve as bioactive peptides with a function independent of protein targeting. A conserved peptide from the signal sequence of major histocompatibility complex (MHC) class I molecules, for example, plays a role in the immune system. The peptide is cleaved by SPP and presented on the cell surface by a non-classical MHC class I molecule termed HLA-E (Figure 2C) (47). Thus this signal sequence-derived peptide controls trafficking and cell surface expression of HLA-E, which allows natural killer immune cells to monitor the integrity of the antigen presenting machinery. Another function of SPP is processing of the hepatitis C virus (HCV) core protein. This nucleocapsid protein is synthesized as the N-terminal portion of a large polyprotein, and upon ER targeting gets released by signal peptidase (48). Subsequently, the immature core protein, which is anchored in the ER membrane by its C-terminus, is further processed by SPP (46,49). The intramembrane cut regulates trafficking of the core protein to cytosolic lipid droplets (49) and viral propagation (50).
In contrast to presenilin, genes encoding homologues of SPP are found in unicellular eukaryotes whose genomes have been analyzed (Table 1) (19,20). This suggests that an SPP-type protease was the ancestral precursor of GxGD-intramembrane proteases. The eukaryotic SPP-like (SPPL) family forms distinct phylogenetic groups. SPP is localized to the ER, whereas several SPPL proteins are localized to other compartments of the secretory pathway (21,22). SPPL2a and SPPL2b have been implicated in the processing of tumor necrosis factor-α (TNFα) (22,51) and processing of a dementia-associated protein termed Bri2 (52). No substrates for SPPL2c and SPPL3 have been reported yet. Genes for SPP and SPPL3 are also found in the malaria parasite Plasmodium falciparum, and inhibitor experiments suggest that the two proteases are important for parasite invasion as well as for growth in human erythrocytes (53,54). The role of SPP-type proteases in P. falciparum and the hepatitis C virus polyprotein processing corroborates that intramembrane proteases are of potential value as therapeutic targets.
Rhomboid Serine Proteases
Rhomboids are intramembrane serine proteases with a catalytic dyad formed by a Ser and His located in two TM domains within a bundle of six conserved TM helices (Figure 3) (55,56). Like in soluble serine proteases, the histidine is believed to strip away the proton from the serine, enabling its nucleophilic attack on the substrate carbonyl group. Rhomboids differ from all the other intramembrane protease families by not requiring prior trimming of the substrate. Because of this striking mechanistic difference, rhomboids can cleave intact membrane proteins and thereby act similarly to sheddases, which release the ectodomian into the extracellular space (57). The first rhomboid was identified by Drosophila genetics as an upstream activator of the epidermal growth factor receptor (EGFR) signaling pathway (58). Rhomboid-1 cleaves the membrane-tethered growth factor precursor termed Spitz, releasing its active form and allowing it to bind EGFR on neighboring cells (Figure 3A) (12,55). The C. elegans rhomboid ROM1 and the mammalian RHBDL2 have similarly been implicated in EGFR control (59,60), whereas the function of the other three rhomboid proteases predicted to act in the mammalian secretory pathway is much less clear. Vertebrate rhomboids form distinct phylogenic groups indicating that they have distinct conserved functions (17,18). Overexpressed mouse rhomboid RHBDL1 has been localized in the Golgi apparatus. In contrast, RHBDL2 and RHBDL3 were predominantly found in the plasma membrane and endosomal structures, respectively (24,60). This diverse localization suggests that the mammalian rhomboids show a spatial selectivity toward potential substrates.
A function that is different from secretion has been described for rhomboids in apicomplexan parasites. In Toxoplasma gondii and P. falciparum, rhomboids contribute to clearance of high-affinity cell surface adhesion molecules during host cell invasion (62,63). Shedding of these adhesion molecules resolves the interaction to the host cell surface and allows the parasite to enter the so-called parasitophorous vacuole. By this shedding event, intramembrane proteases affect trafficking of the entire parasite on its invasive route. Moreover, the rhomboid-released intracellular tail of Toxoplasma adhesion molecules has recently been shown to act as signaling factor, which triggers a shift from the invasive to the replicative phase of the parasite (64). This illustrates how versatile RIP can be used to link a biological transport process to parasite development.
Mitochondrial Rhomboid PARL
A subgroup of rhomboids, named after the human enzyme PARL, is located in the inner mitochondrial membrane (25,26,65). PARL-type rhomboids share the active site motifs with rhomboids in the secretory pathway, but have a matrix targeting sequence and an additional TM domain at the N-terminus (Figure 3B) (17,18). In the budding yeast S. cerevisiae, the PARL-type mitochondrial rhomboid Pcp1 (also termed Rbd1) controls mitochondrial membrane dynamics by cleaving the dynamin-like GTPase Mgm1 (25,26). Mgm1 is anchored in the inner mitochondrial membrane with a TM segment exposing its GTPase domain into the intermembrane space (IMS) (Figure 3B). For its function in membrane fusion, both the membrane-anchored Mgm1 and its cleavage product are required. Rescue experiments have shown that the GTPase activity is dispensable for the Mgm1 precursor (61). This illustrates that RIP also has the potential to convert activity of a membrane-attached protein and not only to trigger its release. How Pcp1 influences the activity of Mgm1 at the molecular level, however, remains to be investigated.
Related but not identical functions in mitochondrial maintenance were assigned to PARL in other species. PARL and its ortholog in D. melanogaster (termed Rhomboid-7), had been linked to apoptosis, type 2 diabetes mellitus and aging (66–68), but the molecular mechanism remains controversial. Initially, it had been suggested that OPA1, the homologue of yeast Mgm1, is cleaved by PARL (66), but several lines of evidence question this original finding (as reviewed in 69). In a number of recent papers, PARL/Rhomboid-7 instead was shown to cleave Omi/HtrA2 (70,71) and Pink1 (71–75), a serine/threonine kinase that has been linked to PD previously (76). The molecular function of Pink1 is still ill defined, but the PARL-catalyzed removal of the N-terminal Pink1 signal anchor sequence has now been shown to determine the fate of the Pink1 kinase domain at the inner mitochondrial membrane. The emerging picture is that PARL can already cleave a Pink1 import intermediate that has only partially passed the translocase of the outer mitochondrial membrane, leading to its reverse translocation into the cytosol and degradation by the ubiquitin proteasome system (72,74,75). More typical of RIP, PARL/Rhomboid-7 also cleaves a subpopulation of Pink1 that has been fully imported and releases its mature form into the IMS (71). There, it confers cell protection against oxidative stress by a so far uncharacterized mechanism (73).
Besides this dual role of PARL in maturation and degradation of Pink1, PARL-catalyzed removal of the Pink1 signal anchor has also been shown to suppress alternative targeting of Pink1 to the mitochondrial outer membrane (75). This apparent mislocalization of the Pink1 precursor occurs in damaged mitochondria and has been shown to drive mitophagy by the recruitment of another PD-associated factor, the E3 ubiquitin ligase Parkin (77–80). It therefore appears that by sampling the efficiency of mitochondrial Pink1 import, PARL-catalyzed removal of the Pink1 signal sequence serves as a checkpoint for mitochondrial integrity (75). Consistent with this idea, a reduction in the mitochondrial mass has been reported upon muscle knockdown of PARL in mice (68) and expression of a cleavage-deficient Pink1 mutant in tissue culture cells (73). The pathophysiology of PARL knockout mice, however, is more complex with secondary processing events probably causing adaptation to the severe accumulation of Pink1 in the outer mitochondrial membrane (72–74). Taken together, there is compelling evidence that PARL is linked to mitochondrial quality control, but more work is needed to unravel the underlying molecular mechanisms.
Regulation of the Intramembrane Cut
Just as with soluble proteins, proteolysis of TM segments needs to be tightly controlled. Uncontrolled protease activity in the plane of the membrane will probably cause severe damage to organelle integrity as had been observed upon ectopic expression of Drosophila Rhomboid-1 in mammalian tissue culture cells (12,55) and overexpression of human PARL (65). Therefore addressing regulatory principles of intramembrane proteolysis is an important research area. The different intramembrane protease families are structurally and evolutionarily unrelated. Nevertheless, all membrane proteases face the same problems in selection of TM substrates such that common regulatory principles have evolved. The first layer of specificity for non-rhomboid intramembrane proteases is substrate tailoring (Figure 2) (28,46,81). Possibly as a consequence of this first shedding event, intramembrane proteases may be able to unfold the substrate TM helix into their active site, a postulated prerequisite for substrate cleavage. In contrast to this conformational control of S2P and GxGD proteases, a recent mutagenesis study combined with structural analysis revealed that rhomboid proteases recognize a defined sequence motif surrounding the scissile peptide bond (82,83). Whether cleavage by other intramembrane proteases is also affected by such primary sequence determinants, however, remains to be addressed. The second, very common principle in RIP is cellular compartmentalization. As outlined in the next paragraph, in the dormant state, the protease and its substrate are located in different sites and cleavage can only occur upon targeting to the same cell compartment (84).
Regulated substrate trafficking: the main layer of control
A dynamic network of targeting and sorting factors controls correct distribution of TM proteins in the eukaryotic secretory pathway (85). SREBP is the first example where regulated trafficking has been shown to control RIP (Figure 4A). Trafficking of the SREBP substrate to the Golgi compartment is controlled in a cholesterol-sensitive manner by the SREBP cleavage-activating protein (SCAP) (86). SCAP binds with high affinity to SREBP in the ER. At high cholesterol levels, the SCAP–SREBP complex interacts with the ER-resident protein Insig-1. This sterol-sensitive interaction prevents transport of SREBP to the Golgi compartment, where S1P and S2P reside (87). However, when cholesterol in the ER membrane is limiting, SCAP does not bind Insig-1 anymore and escorts SREBP to the Golgi compartment for its proteolytic processing and signaling to the nucleus. Another prominent example for transport-dependent RIP is the Rhomboid-1-catalyzed activation of the Drosophila EGFR ligand Spitz (12). Spitz is normally retained in the ER and gets only cleaved by Golgi-resident Rhomboid-1 when a specific transport factor, termed Star, is expressed (12,88). Not surprisingly, RIP in the late secretory pathway is also affected by substrate trafficking. Likewise, γ-secretase-catalyzed cleavage of APP in neurons is affected by sorting factors that target the substrates to specific membrane domains (as reviewed in (84)).
Controlled maturation and trafficking introduce dynamic regulation of γ-secretase activity
A fundamental requirement for specificity is targeting of proteases to the right place. Very little is known about the molecular signals that mediate trafficking of intramembrane proteases. Such signals may not lead to a static distribution of the protease, but could provide an additional layer of regulation. Accordingly, several lines of evidence suggest that dynamic trafficking of γ-secretase regulates processing of APP. This includes Rer1-mediated retrieval of unassembled γ-secretase complexes from the cis-Golgi to the ER (89,90). By interfering with the assembly of nicastrin with Aph-1, Rer1 controls maturation and expression of γ-secretase, whereby Rer1 acts as a negative regulator of γ-secretase-catalyzed APP processing (Figure 4B) (89). Another key event in maturation of presenilin is the autocatalytic cut between the two catalytic TM segments generating a heterodimer of an N-terminal and C-terminal fragment representing the active protease (Figure 2B) (91,92). Thus it appears that, like for classical sheddases and protein convertases, presenilin is synthesized as inactive proform, called zymogen, which is activated by selective endoproteolysis.
Upon maturation of the γ-secretase complex, activity is mainly regulated on the level of substrate tailoring. In addition, trafficking also modulates its activity. Among other sorting factors, β2-adrenergic receptors have been shown to affect endocytosis and activity of the γ-secretase complex (93). Activated β2-adrenergic receptors bind γ-secretase on the plasma membrane, which triggers clathrin-mediated endocytosis and sorting of γ-secretase into late endosomes (Figure 4B). Within the endocytic compartments and the trans-Golgi network, highest rate of APP shedding by β-secretase is observed (94). Taken together with the ligand-induced internalization of γ-secretase, this leads to an increased production of Aβ peptides (93). Furthermore there is evidence that association with specific lipids such as cholesterol affects sorting of γ-secretase thereby modulating its activity (reviewed in (23)).
Expanding diversity: activation of mitochondrial substrates by topology change
A variation of controlled substrate presentation is found in yeast mitochondria, where Pcp1 is regulated by a change of the substrate topology (Figure 3B). Mgm1 is initially integrated in the inner mitochondrial membrane with a bona-fide TM domain, which resists cleavage (25). To allow processing of Mgm1 by Pcp1, this first TM segment needs to be dislocated into the mitochondrial matrix and a second potential TM domain, harboring the rhomboid cleavage site, becomes integrated into the inner mitochondrial membrane (Figure 3B) (95). This second translocation requires ATP, thereby coupling Mgm1 processing to the metabolic state of the cell.
In this review, I have described selected examples of how RIP affects cellular protein trafficking. Remarkably, 13 different intramembrane proteases are scattered across all major cellular organelles in human cells. These proteases have many different functions and there is a corresponding diversity of their molecular action. However, some general themes emerge.
Regulated substrate trafficking and sorting control access of the protease to the substrate. Likewise specific targeting of proteases is central to ensure the selectivity of intramembrane proteolysis. Proteases with the same biochemical properties localize to distinct compartments where they face a specific substrate spectrum. Restricting intramembrane proteases to specific sites includes principles such as zymogen activation, selective retrieval and masking of immature enzymes.
For S2P- and GxGD-type intramembrane proteases, an important layer of specificity is achieved by the strict need for a preceding first cleavage, which is catalyzed by sheddases, protein convertases and signal peptidase. Selectivity of this step is primarily achieved by the juxtamembrane substrate region; the subsequent intramembrane cut typically shows little sequence preference. Rhomboids, on the other hand, cleave close to the membrane surface and therefore they may have evolved higher sequence specificity.
By cleaving within TM anchors, protein ectodomains or peptides are released from cellular membranes. Fragments generated by RIP commonly have a function that is distinct from their precursors. Released cleavage products have different fates ranging from nuclear translocation, selective secretion to detachment from a cell surface adhesion molecule. In addition to triggering protein trafficking, RIP can change the function of a membrane-associated protein, remove a targeting signal or initiate its degradation.
I apologize to all colleagues whose work has not been discussed or cited owing to space limitations. I thank Bernhard Dobberstein and Georg Stöcklin for critical reading of the manuscript. This work is supported by funds from the Baden-Württemberg Stiftung.